[FFmpeg-devel] [PATCH] Use larger tables for yuv > 8 bit to RGB conversion.

Reimar Döffinger Reimar.Doeffinger at gmx.de
Sat Nov 9 20:00:48 CET 2013


On Sat, Nov 09, 2013 at 07:14:35PM +0100, Michael Niedermayer wrote:
> On Sat, Nov 09, 2013 at 04:20:02PM +0100, Reimar Döffinger wrote:
> > This should allow for fairly precise YUV16 to RGB48 conversion
> > for example.
> > However I believe that this specific implementation is not as accurate
> > as it could/should be, i.e. it is buggy.
> 
> also t fails fate:
> fate-qtrle-16bit
> Floating point exception (core dumped)
> 
> 
> > Also the context gets fairly large this way, so I am not sure if
> > this is better than the previous shift-based approach.
> 
> yes
> is support in yuv2rgb for >10 bit needed at all? that is is it common
> enough to optimize for it ?

Probably not worth optimizing for > 10 bit.
However one more problem with changing the tables is that the
normal scaled code uses them, too (libswscale/output.c).
And that code needs the 8-bit ones (at least so far).
And figuring out which will be used at initialization time is messy
if at all possible.
Any ideas?


More information about the ffmpeg-devel mailing list